Cohort Characteristics

In all, 693 patients with MDR pneumonia treated with CAZ-AVI between August 2022 and August 2025 were enrolled (Table 1; Fig. 2). A favorable response was seen in 480 (69.3%)—cure in 356 (51.4%) and improvement in 124 (17.9%). Among the confirmed MDR pathogens, K. pneumoniae was isolated in 54 cases (7.8%), carbapenem-resistant/difficult-to-treat P. aeruginosa (CRPA/DTR-PA) in 171 cases (24.7%), and Acinetobacter baumannii in 13 cases (1.9%). Additionally, mixed polymicrobial infections (isolation of more than one MDR pathogen) were identified in 49 patients (7.1%). It is important to note that while CAZ-AVI lacks intrinsic activity against A. baumannii, the 13 patients whose isolates contained this pathogen received CAZ-AVI either because it was part of a mixed infection with a susceptible Enterobacterales/P. aeruginosa isolate, or because CAZ-AVI was initiated as an empiric regimen prior to the availability of final susceptibility reports.

Table 1 Baseline characteristics by 14-day and 30-day mortality statusFig. 2

Temporal trajectories of inflammatory markers by clinical outcome. n1–n6 denote the landmark cohort sizes for the 14-day non-survivor, 14-day survivor, 30-day non-survivor, 30-day survivor, acute kidney injury (AKI), and non-AKI groups, respectively (n1 = 151, n2 = 520, n3 = 186, n4 = 485, n5 = 14, n6 = 657)

Patients who died by 14 and by 30 days were, compared with survivors, older and more often had malignancy, malnutrition, hypoalbuminemia, CKD or immunosuppressant exposure, and they scored higher on APACHE II, the Sequential Organ Failure Assessment (SOFA) score, and the Charlson index (all P < 0.05). They also more frequently had positive sputum, blood, and urine cultures; bloodstream infection, bacteremia, and urinary tract infection; and septic shock, prior ICU admission, and CRRT. K. pneumoniae, carbapenem-resistant K. pneumoniae (CRKP), A. baumannii, carbapenem-resistant organism (CRO), and CRPA/DTR-PA were each isolated more often from non-survivors (all P < 0.05). Conversely, non-survivors were less likely to have received CAZ-AVI as first-line treatment, more likely to have been given a sub-target maintenance dose, and more likely to display an in vitro non-susceptible phenotype (Table 1).

Univariable Analysis of Mortality

The univariable analysis at 14 and 30 days showed largely consistent findings. Septic shock, prior ICU admission, ventilator-associated pneumonia, urinary tract infection, and a positive urine culture were all associated with higher mortality. At 30 days hospital-acquired pneumonia, bacteremia, and CRRT were also associated with higher mortality. On the microbiological side, CRKP, K. pneumoniae, and A. baumannii were associated with higher mortality, while CRO and difficult-to-treat P. aeruginosa emerged as significant at 30 days. Host characteristics—malnutrition, malignancy, hypoalbuminemia, and a higher APACHE II—showed similar associations. First-line CAZ-AVI, an at-target maintenance dose, combination therapy, and glucocorticoid use, by contrast, appeared potentially protective (all P < 0.05). Such exposure-related associations are vulnerable to confounding by indication and to immortal time bias—patients had to survive long enough to receive first-line CAZ-AVI. Therefore these findings should be interpreted with caution.

Time-Stratified Multivariable Associations

After adjustment, APACHE II remained independently associated with mortality at both 14 days (adjusted OR 1.20; 95% CI 1.14–1.26) and 30 days (1.27; 1.18–1.37), and the variables retained by LASSO agreed with the logistic regression findings (Fig. 3).

Fig. 3

LASSO-penalized regression with cross-validation. a, c deviance versus log(λ) with the minimum and 1-SE criteria; b, d coefficient paths versus log(λ)

For death within 14 days (Fig. 4a), older age and a higher APACHE II aside, bacteremia (4.76; 2.39–9.66), CRO infection (5.93; 2.60–13.60), and septic shock (3.30; 1.32–8.39) were each independently associated with the outcome—in keeping with early dissemination of infection narrowing the window for effective treatment.

Fig. 4

Forest plots of factors independently associated with a 14-day and b 30-day all-cause mortality (multivariable logistic regression)

By 30 days (Fig. 4b), the associated factors were septic shock (5.74; 1.34–24.97) and K. pneumoniae (15.19; 3.25–78.94). The K. pneumoniae estimate sat within a very wide confidence interval—a sign of instability from sparse data—and must be interpreted with caution. To counter small-sample bias we added two sensitivity analyses, Firth penalized logistic regression and a Cox model; these returned an OR of 13.51 (3.65–62.49) and an HR of 10.16 (3.72–27.36), respectively. The three approaches agreed in pointing to a significant positive association between K. pneumoniae and death, and the direction of effect was the same throughout. The point estimate is imprecise given the sample size, but its consistency across models lends the association some credibility.

All analyses of the during- and after-treatment markers and of albumin rested on the pre-specified day 3 landmark cohort (n = 671). Within it, higher IL-6 and IL-10 were associated with worse 30-day outcomes, and a rising serum albumin was associated with lower 30-day mortality (0.89; 0.82–0.95). The landmark design mitigates survivor (immortal time) bias; however, because this is a post hoc observational analysis, residual bias remains possible, and any causal inferences require prospective testing.

Acute Kidney Injury (Safety Observation)

AKI, by the KDIGO definition, occurred in 2.2% of patients (15/693). CRRT use, at 11.4%, was far higher—a gap that points to baseline maintenance dialysis, rather than new AKI, as the main driver of CRRT. Prior ICU admission, septic shock, severe malnutrition, and a positive sputum culture were each more common in the AKI group (Table 2). With so few events, and a rate that sits implausibly low against the baseline CRRT figure, these data can only be considered descriptive. Underascertainment is likely, and they will not support a formal safety conclusion. AKI was therefore not modelled as an independent endpoint.

Table 2 Characteristics stratified by acute kidney injury (AKI) status (descriptive safety observation; AKI was not analyzed as an independent endpoint)Subgroup and Sensitivity Analyses

In patients aged 65 years or older, the associations held: at 14 days for CRO (5.15 [2.01–13.64]) and septic shock (4.96 [1.62–16.69]), and at 30 days for K. pneumoniae (57.14 [11.12–399.38]) and septic shock (12.50 [2.04–121.63]). The K. pneumoniae figure was highly unstable, with an extremely wide confidence interval, and should not be overinterpreted. The Firth and Cox sensitivity analyses gave an OR of 38.25 (15.46–350.78) and an HR of 31.50 (13.55–279.36), again in the same direction. Excluding extreme values left the effect sizes broadly unchanged (Fig. 5).

Fig. 5

Subgroup (≥ 65 years; outlier-excluded) and sensitivity analyses of factors associated with mortality

Inpatient Costs (Descriptive)

Inpatient costs were summarized by bottom-up micro-costing (Sect. “Statistical Analysis”). The median total inpatient cost was USD 14,982 (IQR 8462–25,136), within which CAZ-AVI accounted for a median of USD 3901 (IQR 2069–6159), or about 26% of the pneumonia-related total (Fig. 6). The balance comprised combination or adjunctive antimicrobials, imaging and laboratory tests, single-use consumables, daily bed fees, graded nursing and respiratory support; comorbidity maintenance, non-infection-related investigations, and indirect costs were excluded. These numbers describe the cohort’s resource use and nothing more: we conducted no comparative economic evaluation, computed no incremental cost-effectiveness ratio, and applied no cost-effectiveness threshold (Sect. “Discussion”). Because medians are not additive, the 26% figure should be interpreted only as a rough indication.

Fig. 6

Description of inpatient costs for the cohort (median total inpatient cost and median CAZ-AVI cost)